Motivation




Readmissions networks

Row

Procedures

  • Size represents frequency
  • Thickness represents strength of association (lift)
  • Black represents readmission probability of 1; White, 0

Column

Motivation and Goals

  • Predicting 30-day readmission on the basis of surgery/procedure characteristics
  • Successfully identifying those diagnoses and procedures that can be targeted for quality improvement
  • Hospital Readmissions Reduction Program adjusts payment using an “excess readmission ratio”
  • By choosing the ideal conditions for each procedure, hospitals can increase payment from public insurance providers such as Medicaid and Medicare.

What we have

National Inpatient Sample

ACS NSQIP

  • American College of Surgeons’ National Surgical Quality Improvement Program (NSQIP)
  • High-fidelity data, trained medical billing coders and clinicians ensure accuracy
  • ACS NSQIP identifies more 30-day outcomes than do other quality programs
  • Includes the years 2004 through 2012, about 1.7 million records

Row

Diagnoses

Understanding readmissions




Unplanned readmissions by procedure

  • Yellow indicates more days spent in the hospital for that procedure, while purple indicates fewer days spent in hospital.

Among Hysterectomy Procedures

Readmissions among Hysterectomy Procedures

Readmissions among Hysterectomy Procedures

Detailed characteristics for hysterectomy procedures

Specialty Op Time (min) Days to Discharge % Returned Count
Orthopedics 106.6 1.5 0.056 54
General Surgery 156.4 3.087 0.049 1024
Plastics 140.2 1.781 0.044 228
Urology 219.6 2.982 0.041 170
Gynecology 139.3 1.82 0.037 115278

Modeling readmissions




Readmissions models

partition tree

partition tree

Model performance

Recursive partitioning:

Accuracy: 0.6311 (95% CI: 0.6263, 0.6358)

Random forests:

Accuracy: 0.6287 (95% CI: 0.6191, 0.6383)